Feature selection and classification techniques for effective intrusion detection

Abstract

Security has become a challenging issue due to the rapid growth of number of users in the network. Several Intrusion Detection Systems (IDSs) and Intrusion Prevention Systems (IPSs) have been introduced by various researchers to provide secure and fast communication. This research work has proposed Intrusion Detection System (IDS) techniques with feature selection, classification and outlier detection for improving the detection accuracy. The proposed IDS in this Thesis, uses two new feature selection algorithms, namely, Intelligent Agent and Cuttlefish based Attribute Selection Algorithm (IACASA) and Incremental Feature Selection algorithm (IFSA) which is the combination of IACASA and Extended Chi-Square feature selection algorithm. Existing feature selection algorithm and negative feature selection algorithm (NFSA) are used to enhance the pre-processing activities. The standard data pre-processing activities such as data cleaning, data integration and data transformation are carried out before feature selection. The necessary rules are generated to select the optimal and contributed features from the standard benchmark dataset. An Intelligent Negative Feature Selection Algorithm (INSA) is proposed as part of this research work to improve the classification performance through training. The training stage of this algorithm is separated into initial training and further training. First part of this algorithm covers the non-self-regions and the second part covers the self-regions. In this research work, intelligent agents are used to take effective decisions using fuzzy rules which are framed using the knowledge base information such as facts and rules. Detection accuracy of the proposed IACASA is 99.47% and proposed incremental feature selection algorithm is 99.25%. The proposed INSA provides 98.74% accuracy. newline